kitsunedfox / async-openai

Unofficial Async Python client library for the OpenAI API

async-openai

Unofficial Async Python client library for the OpenAI API based on Documented Specs

Latest Version:

Official Client

Features

• Asyncio based with Sync and Async Support with httpx

• Supports all API endpoints

  • Completions: Docs

  • Edits: Docs

  • Embeddings: Docs

  • Images: Docs

  • Files: Docs

  • Finetuning: Docs

  • Models: Docs

  • Moderations: Docs

  • Search: Docs

• Strongly typed validation of requests and responses with Pydantic Models with transparent access to the raw response and object-based results.

• Handles Retries automatically through backoff

• Supports Local and Remote Cloud Object Storage File Handling Asyncronously through file-io

  • Supports S3: s3://bucket/path/to/file.txt

  • Supports GCS: gs://bucket/path/to/file.txt

  • Supports Minio: minio://bucket/path/to/file.txt

• Supports limited cost tracking for Completions and Edits requests (when stream is not enabled)

---

Installation

# Install from stable
pip install async-openai

# Install from dev/latest
pip install git+https://github.com/GrowthEngineAI/async-openai.git

Quick Usage

import asyncio
from async_openai import OpenAI, settings, CompletionResponse

# Environment variables should pick up the defaults
# however, you can also set them explicitly.

# `api_key` - Your OpenAI API key.                  Env: [`OPENAI_API_KEY`]
# `url` - The URL of the OpenAI API.                Env: [`OPENAI_URL`]
# `api_type` - The OpenAI API type.                 Env: [`OPENAI_API_TYPE`]
# `api_version` - The OpenAI API version.           Env: [`OPENAI_API_VERSION`]
# `organization` - The OpenAI organization.         Env: [`OPENAI_ORGANIZATION`]
# `proxies` - A dictionary of proxies to be used.   Env: [`OPENAI_PROXIES`]
# `timeout` - The timeout in seconds to be used.    Env: [`OPENAI_TIMEOUT`]
# `max_retries` - The number of retries to be used. Env: [`OPENAI_MAX_RETRIES`]

OpenAI.configure(
    api_key = "sk-XXXX",
    organization = "org-XXXX",
    debug_enabled = False,
)

# Alternatively you can configure the settings through environment variables
# settings.configure(
#    api_key = "sk-XXXX",
#     organization = "org-XXXX",
# )


# [Sync] create a completion
# Results return a CompletionResult object
result: CompletionResponse = OpenAI.completions.create(
    prompt = 'say this is a test',
    max_tokens = 4,
    stream = True
)

# print the completion text
# which are concatenated together from the result['choices'][n]['text']

print(result.text)

# print the number of choices returned
print(len(result))

# get the cost consumption for the request
print(result.consumption)

# [Async] create a completion
# All async methods are generally prefixed with `async_`

result: CompletionResponse = asyncio.run(
    OpenAI.completions.async_create(
        prompt = 'say this is a test',
        max_tokens = 4,
        stream = True
    )
)

---

Dependencies

The aim of this library is to be as lightweight as possible. It is built on top of the following libraries:

• aiohttpx: Unified Async / Sync HTTP Client that wraps around httpx

  • httpx: Async / Sync HTTP Requests

  • lazyops: Provides numerous utility functions for working with Async / Sync code and data structures

• pydantic: Type Support

• file-io: Async Cloud-based File Storage I/O

• backoff: Retries with Exponential Backoff